Iterative Statistical Bug Isolation via Hierarchical Instrumentation
No Thumbnail Available
Date
2014-07-14
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Cooperative statistical bug isolation approaches monitor program runtime behavior from end-user executions and apply statistical techniques to pinpoint likely causes of eld failures. Having end-users to run fully instrumented programs is extremely undesirable, as it adds tremendous performance overhead to end-users; it is also unnecessary, since in practice major part of these programs are error-free and need no instrumentation. In order to reduce the monitoring and computational costs as well as the performance overhead of end-users, we propose a novel iterative (and cooperative) statistical bug isolation approach for eld failures via hierarchical instrumentation (HI). Di erent from the existing iterative approaches, our approach via HI does not only maintain small end-user overhead, but also performs iterative bug isolation automatically without developers' intervention, thereby relieving developers of repetitive e orts in isolating bugs. In this paper, we formalize the concept and principles behind the HI technique, and demonstrate its practicality through experimentations. Our experiments show that this new approach via HI saves signi cant instrumentation e ort and sharply reduces runtime overhead, while upholding the accuracy of bug isolation.